Description of problem:

While upgrading the cluster to 4.18 from 4.17, after the cluster network operator finishes upgrading to the 4.18 image, VMs begin to lose network connectivity through their ovn-k8s-cni-overlay localnet NADs.

Restarting the ovnkube-node pod seems to resolve the issue, as does performing a VM live migration for the impacted VM. No OVN DB rebuild was tested as restarting the ovnkube-node pod works.

Running 
ovn-nbctl list logical-switch-port
shows that the impacted VM does not have the logical switch port for connectivity.

This has currently only been brought up for VM pods to my knowledge, I haven't heard it happen with non-VM pods using localnet NADs.

Version-Release number of selected component (if applicable):

4.18.z

Currently has been seen on 4.18.27 and 4.18.28 specifically but is likely wider.

How reproducible:

Currently unsure, has proved difficult so far but I am currently working on a potential reproduction.

Steps to Reproduce:

1. Create a 4.17.z cluster

2. Install and configure OpenShift Virt

3. Configure a ovn-k8s-cni-overlay localnet NAD

4. Create a VM using the NAD configured in step 3

6. Upgrade to 4.18.z

Actual results:

After the cluster network operator upgrades, connectivity is lost to VMs over their localnet NADs.

Expected results:

During and after upgrade connectivity remains for VMs over their localnet NADs.

Additional info:

Details on specific testing will be commented.

Affected Platforms:

OpenShift Container Platform 4.18